Value of machine learning model based on superb microvascular imaging ultrasound radiomics in noninvasive assessment of chronic kidney disease
To noninvasively assess the chronicity of chronic kidney disease (CKD) using a machine learning (ML) model based on superb microvascular imaging (SMI) ultrasound radiomics, and to develop and validate a clinical-radiomics nomogram for individualized prediction. A total of 305 patients with CKD confirmed by renal biopsy were prospectively enrolled and divided into training (n = 213) and test (n = 92) cohorts. Radiomic features were extracted from SMI ultrasound images of the renal cortex. The most valuable features were selected using univariate analysis, Spearman correlation analysis, and the least absolute shrinkage and selection operator (LASSO) algorithm. Five ML algorithms—logistic regression, random forest, support vector machine, k-nearest neighbor, and extreme gradient boosting (XGBoost)—were trained and compared. A clinical-radiomics nomogram was constructed by integrating the radiomics score (Rad-score) with independent clinical predictors. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). The XGBoost model demonstrated the best diagnostic performance among the five ML algorithms, with areas under the ROC curve (AUCs) of 0.903 and 0.888 in the training and test cohorts, respectively. Sex and estimated glomerular filtration rate (eGFR) were identified as independent clinical predictors. The combined clinical-radiomics nomogram, integrating the Rad-score with these clinical factors, showed excellent discrimination, with AUCs of 0.949 and 0.943 in the training and test cohorts, respectively. Calibration curves demonstrated good agreement between predicted and observed probabilities, and DCA confirmed the clinical utility of the nomogram. The ML classifier based on SMI ultrasound radiomics provides a promising noninvasive approach for assessing CKD chronicity. The nomogram combining SMI radiomic features with clinical characteristics offers clinicians comprehensive and personalized imaging information, which may facilitate individualized treatment decision-making.
Authors
- Jiao Yao (ORCID: https://orcid.org/0000-0002-0131-4015)
- Wei Yuan (ORCID: https://orcid.org/0009-0007-5255-3693)
- Jinfeng Qin
- Xiachuan Qin
- Wang Zhou
Institutions
- Sichuan University (CN)
- Chengdu Second People's Hospital (CN)
- First Affiliated Hospital of Anhui Medical University (CN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-19
- DOI
- https://doi.org/10.21203/rs.3.rs-10666978/v1
- Primary Topic
- Radiomics and Machine Learning in Medical Imaging
- Type
- preprint